I think at the current level of LLM code I have observed there's basically zero chance they can produce a competitive cad/cas. Maybe they could approximate an open source kernel like opencascade but I don't see the point in that when freecad already exists.